Measuring the novelty of scientific publications: A fastText and local outlier factor approach

Abstract

Although the novelty of scientific publications has been the subject of previous studies, most have examined the distribution of references in the bibliography, which may not be effective in capturing implied scientific knowledge. We propose an analytical framework for measuring the novelty of scientific publications using a paper's title. At the heart of the framework, fastText is used to construct a vector space model in which papers with similar scientific knowledge are located close to each other, and the local outlier factor is used to measure the novelty of scientific knowledge implied in the papers on a numerical scale. The feasibility and validity of the analytical framework were assessed by comparing the average novelty scores of papers recom-mended with novelty-related tags in Faculty Opinions to those of papers without such tags. This case study of 15,653 papers published in a biomedical journal confirms that our framework is a useful complementary tool for the continuous assessment of the novelty of scientific publications and can serve as a starting point for developing more general models.